Model not supported error while converting Yolov7 weights 'IDetect layer'

[junaidp1]

I am getting Model not supported error when I use gen_wts_yoloV7.py to generate weights and config file. Yolov7 model that I have contains IDetect layer which I don't see in the gen_wts_yoloV7.py.

Any idea how to fix it? Thanks

[marcoslucianops]

The IDetect layer isn't implemented yet. I will add soon.

[GiorgosBetsos]

I also have a custom Yolo v7 model and get the same error when trying to convert using gen_wts_yoloV7.py. How can I tell if the error is due the same reason described here, or due to some other reason?

UPDATE:

Ok, I have modified gen_wts_yoloV7.py to print name of the model, it is also IDetect layer in my case.

[ksd777]

When can we expect implementation of IDetect layer? Or maybe you can let me know which version of Yolov7 is using Detect layer instead of IDetect?

[franckmarineai]

IDetect support is what we are waiting for to complete an important project. Any ETA for that ?

[rbgreenway]

Need IDetect support badly.
I'm assuming that there are yolov7 models that don't use IDetect. Which ones? At least need an interim solution.

[emirkiani]

We have the same error while trying to convert the custom yolov7 model.

[marcoslucianops]

The IDetect layer is used in re-parameterized models. I will work on it in the next days.

[pullmyleg]

@marcoslucianops I dont think you need to support idetect :). From my understanding Idetect is for training, detect is for inference (faster) as per the scripts in the cfg/deploy folder of yolov7. I just got this working with V7 tiny using your current tools.

This has been tested with Deepstream 6.0.1 running on a jetson nano.

Steps:

1. Reparamitise the model. Note you need to update the reparamitizer script replacing 105 with 77 if you are using yolov7-tiny or another unsupported model.
2. use the config from cfg/deploy supplied in repo.
3. convert as per usual using existing tooling.

[marcoslucianops]

Yes, the IDetect is for training. You need to do the reparameterization to convert the model using the gen_wts_yoloV7.py.

[animeesh]

https://colab.research.google.com/drive/17HIHsKAdfTbZQSu7LxZIwtHQr7LZYSKD?usp=sharing this colab file can help you guys in .pt to .wts model conversion

[anchalgupta2412]

Weights obtained from re-parameterization differ from original weights. I am getting different recall & precision on same images with 2 models.

[marcoslucianops]

Weights obtained from re-parameterization differ from original weights. I am getting different recall & precision on same images with 2 models.

Did you test using YOLOv7 repo?